iasinDev / skforecast

Time series forecasting with scikit-learn models

skforecast

Time series forecasting with scikit-learn regressors.

Skforecast is a python library that eases using scikit-learn regressors as multi-step forecasters. It also works with any regressor compatible with the scikit-learn API (pipelines, CatBoost, LightGBM, XGBoost, Ranger...).

Documentation: https://joaquinamatrodrigo.github.io/skforecast/

Installation

pip install skforecast

Specific version:

pip install skforecast==0.4.3

Latest (unstable):

pip install git+https://github.com/JoaquinAmatRodrigo/skforecast#master

The most common error when importing the library is:

'cannot import name 'mean_absolute_percentage_error' from 'sklearn.metrics'.

This is because the scikit-learn installation is lower than 0.24. Try to upgrade scikit-learn with:

pip3 install -U scikit-learn

Dependencies

• numpy>=1.20, <=1.22
• pandas>=1.2, <=1.4
• tqdm>=4.57.0, <=4.62
• scikit-learn>=1.0, <=1.1.1
• statsmodels>=0.12, <=0.13.2
• optuna==2.10.0
• scikit-optimize==0.9.0

Features

• Create recursive autoregressive forecasters from any regressor that follows the scikit-learn API
• Create multi-output autoregressive forecasters from any regressor that follows the scikit-learn API
• Grid search to find optimal hyperparameters
• Grid search to find optimal lags (predictors)
• Include exogenous variables as predictors
• Include custom predictors (rolling mean, rolling variance ...)
• Multiple backtesting methods for model validation
• Include custom metrics for model validation
• Prediction interval estimated by bootstrapping
• Get predictor importance

What is coming in the new release?

• Random search and bayesian search (using optuna or skopt) for hyperparameter optimization. Try random_search_forecaster() and bayesian_search_forecaster().
• ForecasterAutoregMultiOutput has been renamed to ForecasterAutoregDirect.
• Modeling multiple time series simultaneously. Try ForecasterAutoregMultiSeries and the new functions availables in model_selection_multiseries.
• Allow different transformations for each predictor (lags and exogenous).
• Bug fixes and performance improvements.

Try it:

pip install git+https://github.com/JoaquinAmatRodrigo/skforecast#master

Visit changelog to view all notable changes.

Documentation

The documentation for the latest release is at skforecast docs .

Recent improvements are highlighted in the release notes.

• Introduction to time series and forecasting

• Recursive multi-step forecasting

• Backtesting (validation) of forecasting models

• Grid search of forecasting models

• Prediction intervals

• Using forecaster in production

Examples and tutorials

English

• Skforecast: time series forecasting with Python and Scikit-learn

• Forecasting electricity demand with Python

• Forecasting web traffic with machine learning and Python

• Forecasting time series with gradient boosting: Skforecast, XGBoost, LightGBM and CatBoost

• Bitcoin price prediction with Python

• Prediction intervals in forecasting models

Español

• Skforecast: forecasting series temporales con Python y Scikit-learn

• Forecasting de la demanda eléctrica

• Forecasting de las visitas a una página web

• Forecasting series temporales con gradient boosting: Skforecast, XGBoost, LightGBM y CatBoost

• Predicción del precio de Bitcoin con Python

• Workshop predicción de series temporales con machine learning Universidad de Deusto / Deustuko Unibertsitatea

• Intervalos de predicción en modelos de forecasting

Donating

If you found skforecast useful, you can support us with a donation. Your contribution will help to continue developing and improving this project. Many thanks!

Citation

If you use this software, please cite it using the following metadata.

APA:

Amat Rodrigo, J., & Escobar Ortiz, J. skforecast (Version 0.5.0) [Computer software]

BibTeX:

@software{skforecast,
author = {Amat Rodrigo, Joaquin and Escobar Ortiz, Javier},
license = {MIT},
title = {{skforecast}},
version = {0.5.0}
}

View citation file.

License

joaquinAmatRodrigo/skforecast is licensed under the MIT License, a short and simple permissive license with conditions only requiring preservation of copyright and license notices. Licensed works, modifications, and larger works may be distributed under different terms and without source code.